Institute of research into the fundamental laws of the Universe

Galaxy clusters, formed at the intersection of the filamentary large scale structures are the most visible tracers of the distribution of matter in the Universe. Composed of dark matter (85% of the total mass) and baryons (hot X-ray emitting gas and galaxies), the evolution of the cluster population over time gives insight on the structure formation and reflects the underlying cosmology.

The sensitivity of Euclid, the upcoming major cosmological mission led by the European Space Agency and planned to be launched in 2023, should allow blind detection of clusters through their weak lensing signal i.e. the coherent distortion of background galaxy images by the intervening mass of the cluster, an effect directly linked to their total (dark and baryonic) projected mass. Combined with the sky coverage, this will allow the construction of a significant galaxy cluster catalogue that is for the first time truly representative of the true cluster population. Indeed, up to now all galaxy cluster catalogues rely on detection through their baryonic signal that represents only 15% of the total mass (e.g. through the intra-cluster gas content in X-rays and the Sunyaev-Zeldovich effect (SZE) at millimetre wavelengths, or through the optical light in the galaxies). The catalogue of galaxy clusters detected through weak lensing is directly linked to the total mass of the cluster. This will provide new constraints on galaxy cluster abundances in the Universe, which has important implications for cosmology.

The thesis project aims at development of innovative methods to detect and characterise clusters through their weak lensing signal. With the imminent launch of the Euclid satellite, the thesis project will take place in a very stimulating context. The ultimate goal of the project being to apply the methods to Euclid data and to take part in the scientific exploitation.

The LISA space observatory, scheduled for launch in 2035, will consist of three satellites 2.5 million kilometers apart and will allow the direct detection of gravitational waves undetectable by terrestrial interferometers, opening a new window of observations in astrophysics. In order to maximize the scientific potential of such a mission, the analysis of the data will involve several steps, one of which is the rapid analysis pipeline, whose role is the detection of new events, as well as the characterization of events. Beyond the interest for LISA, this low latency analysis pipeline plays a key role for the fast follow-up of events detected by electromagnetic observations (ground or space observatories, from radio waves to Gamma rays). If fast analysis methods have been developed for ground-based interferometers, the case of space-based interferometers such as LISA remains a field to explore. Thus, an adapted data processing will have to take into account the mode of transmission of the data by packet, thus requiring the detection of events from incomplete data, marred by artifacts. These methods will have to allow the detection and characterization of events as diverse as black hole mergers, EMRIs (extreme mass ratio inspirals), bursts and binaries of compact objects. All this must be done in real time. To this end, this thesis will aim at generalizing classical methods, based on matched filtering, to the analysis of LISA data and at developing a new approach based on machine learning for the detection and early characterization of black hole mergers. These methods will be done in the framework of the LISA consortium and will contribute to the development of a fast analysis pipeline in France.

Blind and semi-blind unmixing problems are classical inverse problems and ubiquitous in a very wide range of scientific domains from sound processing, medical signal processing to remote sensing or astrophysics. In these domains, the fast development of high resolution/high sensitivity multispectral sensors mandates the development of dedicated analysis tools. For such type of data, the observations can be modelled as the linear or non-linear combination of various elementary physical components, which are to be retrieved by the astrophysicist. However state-of-art methods suffer two major bottlenecks when facing real-world applications: i) the ability to retrieve physically interpretable solutions, ii) their high computational cost, which largely limit their applicability. To that end, the objective of this work is to investigate new approaches, based on machine learning, to tackle blind and semi-blind (when one has access to no or only partial knowledge about the components to be restored) unmixing problem. More precisely, we introduced a novel algorithm based on unrolling techniques to tackle supervised unmixing. We showed that unrolling techniques allow to account for physics-driven information to unmix the data, leading to more physically relevant solutions at a very low computational cost. The goal is then to generalize this prior work to the more challenging blind/semi-blind cases, which will require both revisiting the model architecture as well as optimisation. The results will be tested and validated with astrophysical X-ray data (e.g. Chandra) as well as gravitational wave simulations in preparation for LISA.

Future experiments on linear colliders (e+e-) with low hadronic background require improvements in the spatial resolution of pixel vertex detectors to the micron range, in order to determine precisely the primary and secondary vertices for particles with a high transverse momentum. This kind of detector is set closest to the interaction point. This will provide the opportunity to make precision lifetime measurements of short-lived charged particles. We need to develop pixels arrays with a pixel dimension below the micron squared. The proposed technologies (DOTPIX: Quantum Dot Pixels) should give a significant advance in particle tracking and vertexing. Although the principle of these new devices has been already been studied in IRFU (see reference), this doctoral work should focus on the study of real devices which should then be fabricated using nanotechnologies in collaboration with other Institutes. This should require the use of simulation codes and the fabrication of test structures. Applications outside basics physics are X ray imaging and optimum resolution sensors for visible light holographic cameras.

After a very successful operation period crowned with the discovery of the Higgs boson, the Large Hadron Collider (LHC) will undergo a luminosity upgrade where it is planned to increase the collision rate by a factor of ten, resulting with extremely large number of simultaneous collisions. The particle detectors of the LHC will also be upgraded to cope with these challenging environment. Furthermore, with increased granularity and more advanced readout electronics, they are aimed to achieve a better event reconstruction for instance with new high-granularity calorimeters.



In this project we will develop an ultra-low latency; machine learning based electromagnetic or hadronic particle reconstruction algorithm for the collider experiments. This state-of-the-art algorithm will be distributed into large number of high-capacity low-latency components, which will drastically improve the readout efficiency and reconstruction capability of the future collider experiments. This unified algorithm will have significant impact on the ambitious physics program of these colossal detectors. We will demonstrate the impact of this development for Higgs precision measurements, focusing on the analysis of the Higgs self coupling.



Implementing advanced machine learning algorithms in low-level electronics such as Field Programmable Gate Arrays (FPGA) is a newly-emerging exciting field. To accomplish the goals of the project we will be collaborating with other international laboratories and institutes such as CERN, Fermilab, CalTech, with frequent visits to these labs. The successful candidate will be working with the High Level Synthesis tools to optimize the neural networks to their limits. They will need to have basic programming knowledge on C++ and python, and some exposure to readout systems will be a plus.